TDX attestation 403 on Ubuntu 26.04 with RTX PRO 6000 (QEMU 10.2.1)

[WinstonFrankBot]

Problem

TEE VM attestation fails with HTTP 403 "Authentication failed" when running on Ubuntu 26.04 LTS (kernel 7.0.0-27-generic) with RTX PRO 6000 Blackwell Server Edition GPUs.

Environment

• OS: Ubuntu 26.04 LTS (Resolute Raccoon)
• Kernel: 7.0.0-27-generic
• QEMU: 10.2.1 (Debian 1:10.2.1+ds-1ubuntu3.1)
• GPU: 8x NVIDIA RTX PRO 6000 Blackwell Server Edition
• Guest Image: SHA256 2df7256a248b246ae532b74601e376a455ad8af98202e56deb91b623fce3b88a (matches expected)
• sek8s: Latest from main branch
• PCCS: Running with valid Intel API key
• TDX: Active (kvm_intel tdx=Y)
• Topology: NumaTopology(gpu_nodes=(0, 0, 0, 0, 1, 1, 1, 1)) — matches baselined profile

Error

TDX base: https://tdx-attestation.chutes.ai:8443, validator base: https://api.chutes.ai
LUKS key fetch attempt 1/3
Begin: Fetching nonce ... done.
Begin: Generating TDX quote ... Quote generated: 5243 bytes, saved to /tmp/tdx_quote.bin
Begin: Sending attestation for VM 'chutes-miner-tee-0' ... done.
Failure: Authentication failed (HTTP 403)

Analysis

• TDX quote generates successfully (5243 bytes)
• PCCS is configured and running with Intel PCS API key
• Guest image hash matches exactly
• QEMU 10.2.1 is baselined in the RTX PRO 6000 profile
• The sek8s RTX PRO 6000 profile has a local baseline for QEMU 10.2.1 but Ubuntu 26.04 is listed as "not yet end-to-end validated" in the topology matrix
• Ubuntu 25.10 (kernel 6.11) is validated but the generic 6.11 kernel doesn't have TDX support compiled in

Request

Please baseline Ubuntu 26.04 (kernel 7.0, QEMU 10.2.1) RTMR measurements for RTX PRO 6000 in the remote attestation server, or provide guidance on getting Ubuntu 25.10 TDX-enabled kernel running on a 26.04 install.

Miner Info

• SS58: 5FWMNePMuMTiPRxwYmRamQHqUeo1B5JXUbBoQZYC8zUSoqxJ
• UID: 107 on SN64

[WinstonFrankBot]

Update — Root Cause Identified

The 403 is NOT caused by the Ubuntu version or kernel. It's a NUMA topology mismatch.

Our server (Intel Xeon 6767P Granite Rapids, dual-socket) has Sub-NUMA Clustering (SNC) enabled in BIOS, resulting in 4 NUMA nodes. GPUs are on nodes 0 and 2.

The sek8s RTX PRO 6000 profile locally accepts both:

• NumaTopology(gpu_nodes=(0, 0, 0, 0, 1, 1, 1, 1)) — 2 NUMA nodes
• FlatTopology(gpu_count=8) — flat fallback for >2 NUMA nodes

Our host fingerprint is FlatTopology(gpu_count=8, nvswitch_count=0, ib_count=0) which matches the local baseline. But the remote attestation server at tdx-attestation.chutes.ai:8443 returns 403, suggesting it only has the NumaTopology RTMR0 value whitelisted, not the FlatTopology one.

Request: Please add the FlatTopology(gpu_count=8) RTMR0 measurement for RTX PRO 6000 with QEMU 10.2.1 to the remote attestation allowlist. The sek8s code already considers this a valid topology.

We cannot disable SNC via OS — the AMI BIOS overrides EFI variable changes on reboot, and this is a managed data center server without direct BIOS console access.